Method for accurately predicting the wall cutting feed force of a cutting wheel when a shield tunneling machine passes through a reinforced concrete diaphragm wall

Abstract

The present invention discloses a method for accurately predicting the wall-cutting feed force of a cutterhead when a shield tunneling machine passes through a reinforced concrete diaphragm wall, comprising the following steps: (1) collecting data from actual construction projects and field tests, including the operating parameters of the shield tunneling machine during its passage through the reinforced concrete diaphragm wall; and performing preprocessing; (2) analyzing the data from actual construction projects and field tests to reveal a mapping relationship between the total shield tunneling feed force and the wall-cutting feed force of the cutterhead; (3) creating a BO-BiGRU-Attention hybrid prediction model to predict the total shield tunneling feed force;(4) Applying the model to actual construction projects, whereby the wall cutting feed force of the cutting wheel is predicted in real time by combining the mapping relationship with the predictive model. The present invention can predict the wall cutting thrust of the cutting wheel in real time and provide accurate instructions for construction.

Description

2 typically oversimplifications. They do not optimally utilize the extensive information contained in the monitoring data, which leads to limited prediction accuracies. Disclosure of the Invention: The objective of the present invention is to provide a method for accurately predicting a wall-cutting feed force of a cutter wheel when a shield tunneling machine passes through a reinforced concrete diaphragm wall, which solves the problem of difficult data acquisition on the construction site. Technical Solutions of the Invention: A method for accurately predicting a wall-cutting feed force of a cutter wheel when a shield tunneling machine passes through a reinforced concrete diaphragm wall according to the present invention, comprising the following steps: (1) Collecting data from actual construction projects and field tests, including the operating state parameters of the shield tunneling machine while passing through the(1) Shield tunneling machine through the reinforced concrete diaphragm wall; and performing pre-processing; (2) Analyzing data from actual construction projects and field tests to reveal a mapping relationship between a shield tunneling total feed force and the wall cutting feed force of the cutter wheel; (3) Creating a BO-BiGRU-Attention hybrid prediction model to predict the shield tunneling total feed force; (4) Applying the model to actual construction projects, combining the mapping relationship with the prediction model to predict the wall cutting feed force of the cutter wheel in real time. Furthermore, in step (1) the parameters specifically include feed rate, cutting wheel rotational speed, shield feed total feed force, and wall cutting feed force of the cutting wheel; the preprocessing is specifically carried out by treating an outlier value by mean value replacement, where the formula is as follows: BE2025 / 5428 3 2 i-1,ji+1,ji,j y+yy= wherei,jyrepresents the outlier value, represents the row number of the outlier value, and represents the column number of the outlier value. Furthermore, step (2) comprises the following steps: (21) Constructing a coordinate system wherein a positional relationship between the cutterhead of the shield tunneling machine and the diaphragm wall can be subdivided into phases ny1, y2, and y3, where B is the thickness of the diaphragm wall and D is the diameter of the cutterhead; (22) Determining the actual number of tools cutting simultaneously during an entire wall cutting operation using a model wherein, by combining with the positional coordinates of roller cutters, a time-dependent progression of the number of roller cutters is obtained, and after performing a Gaussian fit, the continuous, precise model for the time-dependent change in the number of tools is generated; (23) Calculating the ratio of the wall cutting feed force of the cutter wheel to the shield feed total feed force using the following formula: 15 =++ = / =(πD / 4−) =+ 1112211 / FkFFFFwhere feed force data at constant feed rate (mm / min) are taken from field tests; the other parameter values ​​are taken from actual construction projects: n – number of tools acting on the wall at this time, pieces; N – total number of tools; – average contact area between the cutter wheel and the diaphragm wall during the cutting process, m²; L – length of the shield tunneling machine, m; aP – earth pressure, bar; c – reduction of the uniaxial compressive strength of MJS, MPa; α – reduction factor; 1 μ – friction factor between MJS layer and shield tunneling shell; η – opening ratio of the cutter wheel; – total weight of the trailing equipment of the shield tunneling machine, t; Fk – ratio coefficient between cutter wheel feed force and shield tunneling feed force. Furthermore, step (3) includes the following steps: 5 (31) Normalizing data, where the formula is as follows: xnormalized = x−xMin xMax−xMin where xnormalized is the value of the normalized shield feed total feed force(32) Feature selection: Selecting key features with high correlation to the total shield feed force from the preprocessed data by Spearman correlation analysis; and splitting into a training set and a validation set; (33) Creating a BO-BiGRU-Attention hybrid prediction model, including: input layer, BiGRU layer, attention layer, Bayesian optimization, and output layer. Furthermore, the BiGRU layer is specifically designed to capture temporal dependencies in the data using a bidirectional engaged recurrent unit network. Furthermore, the attention layer is specifically designed such that an attention mechanism is applied to the output layer of the BiGRU layer, which enables the model to calculate key information by determining the weighting of each time step.to gain from the data.25 Furthermore, Bayesian optimization comprises the following steps: S1 Initialization: Selecting an initial set of sample points; S2 Construction of a probability model: Creating a probability model using Gaussian regression based on the available sample points; S3 Selection of a next sample point: Drawing a conclusion using the probability model and selecting the next sample point based on the results of the conclusion;5 S4 Evaluation of an objective function MAE at the selected sample point to obtain a new observation value; S5 Updating the probability model: Adding a new sample point and observation value to the existing sample point set and updating the probability model; 10 S6 Checking a termination condition: Determining whether the algorithm should be terminated, based on whether the preset objective function converges, and ona maximum number of iterations. Furthermore, the output layer is specially designed such that, after denormalization of the data, a predicted value of the total shield tunneling feed force is output and the performance of the model is evaluated using a loss function, where the formulas are as follows: the denormalization formula is as follows: x = x normalized (xMax − xMin) + xMin; the formula for calculating the loss value is as follows: 20 = 1 (′ −) An electronic device comprising a memory, a processor, and a computer program stored in memory and executable on the processor, wherein the computer program, when loaded into the processor, implements a method for accurately predicting a wall cutting feed force of a cutter wheel when a shield tunneling machine passes through a reinforced concrete diaphragm wall according to one of the above embodiments. A storage medium on which my computer program is stored, wherein the BE2025 / 5428 6 computer program, when executed by a processor, is a method forThe accurate prediction of a wall-cutting feed force of a cutter wheel when a shield tunneling machine passes through a reinforced concrete diaphragm wall is implemented according to one of the above embodiments. Advantageous effects: Compared to the prior art, the present invention has the following essential advantages: improved data utilization; discovery of the mapping relationship between the total shield tunneling feed force and the wall-cutting feed force of the cutter wheel, thereby creating a theoretical basis for accurate predictions; the created BO-BiGRU-Attention hybrid prediction model exhibits high prediction accuracy and enables real-time prediction of the wall-cutting feed force of the cutter wheel, thereby providing accurate instructions for construction; improved construction safety and efficiency of the shield tunneling machine passing through the reinforced concrete diaphragm wall, and reduced construction costs. Brief description of the drawings 15Fig. 1 shows a coordinate system created based on the positional relationship between the cutterhead surface of a shield tunneling machine and a diaphragm wall according to the present invention. Fig. 2 shows the theoretical change in the number of tools and the Gaussian adjustment curve during the wall cutting process according to the present invention. Fig. 3 shows the proportion of the wall cutting feed force to the total feed force of the present invention. Fig. 4 is a Spearman correlation analysis diagram of the present invention. Fig. 5 is a diagram of the BiGRU structure of the present invention. Fig. 6 is a schematic diagram of the attention structure of the present invention. Fig. 7 is a flowchart of the Bayesian optimization steps of the present invention. Fig. 8 is a flowchart of the BO-BiGRU attention algorithm of the present invention. BE2025 / 5428 7 Fig.9 is a density scatter diagram of the actual and predicted values ​​of the shield drive total feed force according to the present invention.Fig. 10 shows error values ​​of the feed force in the training set and in the validation set of the present invention. Fig. 11 is a density scatter plot of the theoretical and predicted values ​​of the 5 wall cutting feed force of the cutting wheel of the present invention. Detailed embodiments In the following, a further description of the technical solution of the present invention is given in conjunction with the drawings. As shown in Fig. 1, embodiments of the present invention constitute a method 10 for the accurate prediction of a wall-cutting feed force of a cutter wheel when a shield tunneling machine passes through a reinforced concrete diaphragm wall, comprising the following steps: (1) collecting data from actual construction projects and field tests, including the operating condition parameters of the shield tunneling machine during passage of the shield tunneling machine through the reinforced concrete diaphragm wall; and performing a preprocessing procedure, in which the parameters specifically include the advance rate and rotational speed of the cutter wheel.The cutting wheel, shield feed total feed force and wall cutting feed force of the cutting wheel include; the preprocessing is carried out specifically by treating an outlier value by mean value replacement, where the formula is as follows: 20 2 i-1,ji+1,ji,j y+yy= where i,jy represents the outlier value, idi represents the row number of the outlier value and j represents the column number of the outlier value. (2) Analyzing data from actual construction projects and field trials to reveal a mapping relationship between the total shield drive feed force and the 25 wall cutting feed force of the cutter wheel, comprising the following steps: BE2025 / 5428 8 (21) Constructing a coordinate system as shown in Fig. 1, wherein a positional relationship between the cutter wheel of the shield tunneling machine and the diaphragm wall can be divided into phases y1, y2 and y3, where y is the thickness of the diaphragm wall and y is the diameter of the cutter wheel; (22) Determining the actual number of tools cutting simultaneously during a 5the entire wall cutting process using a model as shown in Fig. 2, whereby by combining it with the position coordinates of roller bits a time course of the number of roller bits is obtained, and after performing a Gaussian fit the continuous, precise model for the time change of the number of tools is generated; 10 (23) Calculating the ratio of the wall cutting feed force of the cutter wheel to the total shield feed force of the following formula: =++ = / =(πD / 4−) =+ 1112211 / FkFFFF where feed force data at the same feed rate (mm / min) are taken from field tests; the other parameter values ​​come from actual construction projects: n – number of tools on the wall at this time 15 act, piece; N – total number of tools; – average contact area between the cutting wheel and the diaphragm wall during the cutting process, m²; L – length of the shield tunneling machine, m; aP – earth pressure, bar; c – reduction of the uniaxial compressive strength of MJS, MPa; α – reduction factor; 1 μ –Friction factor between MJS – layer and shield tunneling shell; η – cutterhead opening ratio 20; – total weight of the trailing equipment of the tunneling machine, t; Fk – ratio coefficient between cutterhead feed force and shield tunneling feed force. The parameters are as shown in Table 1: BE2025 / 5428 9 Feed rate 11 F / kN 12 F / kN 2 F / kN / kN / kN Error Fk 12 193.84 1699.96 7938.29 11832.09 10982.45 7.74% 18.54% 1.52 264.01 1727.23 7938.29 11929.53 12303.50 3.04% 18.98% 22300,711736,327938,2911975,3313406,8310,68%19,21% 2,52515,001727,237938,2912180,5213823,5211,89%20,65% 32969,311772,687938,2912680,2914979,0015,35%23,42% The ratio of the wall cutting feed force of the cutting wheel to the shield drive total feed force is as shown in Fig. 3. (3) Creating a BO-BiGRU attention hybrid prediction model for predicting the shield tunneling total feed force, comprising the following steps: (31) Normalizing data, the formula being as follows: 5 xnormalized = x−xMin xMax−xMinwhere xnormalized represents the value of the normalized shield drive total feed force, x represents the original value of the shield drive total feed force, xMax represents the maximum value of the shield drive total feed force, xMin represents the minimum value of the shield drive total feed force; (34) Feature selection: Selecting key features with high correlation to the shield drive total feed force from the preprocessed data by Spearman correlation analysis, as shown in Fig. 4; and splitting into a training set, a validation set and a test set; (35) Creating a BO-BiGRU-Attention hybrid prediction model, including: input layer, BiGRU layer, attention layer, Bayesian optimization and output layer, as shown in Figures 5 to 8. The BiGRU layer is specifically designed to detect temporal dependencies in the data using a bidirectional engaged recurrent unit network.20 The attention layer is specifically designed such that the output layer of theBiGRU layer employs an attention mechanism that allows the model BE2025 / 5428 10 to extract key information from the data by calculating the weighting of each time step. Bayesian optimization comprises the following steps: S1 Initialization: Selecting an initial set of sample points; S2 Probability Model Construction: Constructing a probability model using Gaussian regression based on the available sample points; S3 Next Sample Point Selection: Drawing a conclusion using the probability model and selecting the next sample point based on the results of the conclusion; S4 Evaluation of an objective function MAE at the selected sample point to obtain a new observation value; S5 Updating the probability model: Adding a new sample point and observation value to the existing sample point set and updating the probability model; 15S6 Checking a termination condition: Determine whether the algorithm should be terminated, based on whether the preset objective function converges, and on a maximum number of iterations. The Bayesian optimization ultimately determines the optimal hyperparameters of the model, as shown in Table 2: 20 Model Number of neurons Number of iterations Learning rate Loss rate Number of network layers BO-Bi-Attention 87 2000.00 80.19 2 The output layer is specifically designed such that, after denormalization of the data, a predicted value of the total shield thrust force is output and the performance of the model is evaluated using a loss function, where the formulas are as follows: the denormalization formula is as follows: 25 x = xnormalized(xMax − xMin) + xMin BE2025 / 5428 11 the formula for calculating the loss value is as follows: = 1 (′ −) The distribution of relative positions between the actual total thrust forces and the predicted total thrust forces are shown in Fig. 9. Fig. 10 shows theError history of the model developed in the present invention for data set 5 of feed force and torque. From the figures, it is evident that the error values ​​on the training and validation set gradually decrease with increasing iteration number and eventually tend to stabilize. When the iteration number is increased to 200, the prediction error of the feed force drops to a very low and stable level. The final error values ​​on training and validation set 10 are 0.002 and 0.003, respectively. The low error values ​​show that the prediction model proposed in the present invention avoids the problem of overfitting and has good prediction accuracy and generalization capability. (4) Applying the model to the actual construction projects, whereby, by combining the mapping relationship with the prediction model, the wall cutting feed force of the 15 cutting wheel is predicted in real time. Specifically, the concrete strategy consists of correlating the ratio value with the advance rate. If theIf the propulsion speed lies within a certain range, a corresponding ratio coefficient is assigned, which allows the ratio coefficient to be dynamically selected to increase the accuracy of the model, as in